<![CDATA[The rapid growth of e-commerce platforms in Indonesia has heightened the need to understand customer sentiment for improving service quality and user experience. This study conducts sentiment analysis on customer reviews from two e-commerce platforms under the Baliyoni Group: Tokodaring.balimall.id and Balimall.id. A total of 45,257 reviews were collected through database dumps, consisting of 41,519 reviews from Tokodaring and 3,738 from Balimall. The raw textual data underwent a comprehensive preprocessing pipeline involving duplicate removal, tokenization, normalization, stopword elimination, stemming, and sentiment labeling. Two classification models were implemented and compared: Gated Recurrent Unit (GRU), a deep learning model that captures sequential dependencies, and Naïve Bayes, a probabilistic classifier utilizing TF-IDF features. Evaluation results indicate that the GRU model achieved higher precision and F1-score, whereas Naïve Bayes obtained perfect recall but lower precision. Specifically, for Balimall reviews, Naïve Bayes achieved an accuracy of 90.06%, precision of 84.03%, recall of 100%, and an F1-score of 91.32%. These results mirror the Tokodaring dataset performance, confirming the model’s ability to effectively identify positive sentiment but also its tendency to overestimate it, leading to false positives on negative reviews. Overall, the GRU model demonstrated a more balanced performance, making it more suitable for analyzing informal and diverse user-generated text on Indonesian e-commerce platforms. This research provides insights into developing robust sentiment analysis systems to support data-driven decision-making in digital commerce environments.]]>
